Hi all, I built a multi-document agent

its just using a vector store index. You might have to adjust the top k, or even use a reranker

Thank you for your quick reply @Logan M. I realize it's a vector store index, but my question is what is being indexed? The QueryEngineTool that each document agent is embedded in has name and description in its metadata, and I though this is what is being indexed and searched. But even in simple cases where the name is "Document_1", the retriever does not return the right agent when the query is just "Docuent_1".

This is the function for creating the nodes to be embedded

So basically the name and description

If the names and descriptions are super generic (document_1, document_2), it will not work very well

Thank you @Logan M , that makes sense. For my use case it will be very common for users to request information using simple document names. I am going to experiment with using an AutoRetriever that filters using metadata.